but so does e.g. the Sun Position Addon [and there it is working properly].

that does not do the following though

It is caused by rB08dbc4f996e4: PyAPI: use postponed annotations to support Python 3.10

• Add assert if any *unsafe* tags are set as extra tags.

• Don't set session UUID for temporary ID's

Use methods to avoid passing the BlendData as an argument to 'load'.

• Correct comment

LGTM except for one little thing.

Cannot reproduce here (tested in both 2.91 and current 2.93)

In T85841#1121668, @Bastien Montagne (mont29) wrote:

@Dylan Reeve (cadtofu) please do not set release goals yourself, only active developers should be in charge of that.

Looks good. Two changes I will make when committing:

• Comment formatting: Comments should start with a capital letter and end with a period.
• Remove the "for python access" comment from the header file (since it's not quite true anymore) and move the prototype up a bit to be with similar functions.

Didn't first think of changing the python resetview, but for consistency it got changed as well to reset_view.

Adjusted the naming as per suggestion by @Hans Goudey (HooglyBoogly).

Makes sense to me, just a couple of small comments.

@Dylan Reeve (cadtofu) please do not set release goals yourself, only active developers should be in charge of that.

Maybe @Hans Goudey (HooglyBoogly) can give green light?

@Germano Cavalcante (mano-wii) Thank you for looking into this. Indeed the callback itself removes handler, causing the issue you've identified.

Thank you for having a look at this.
I also thought about this. This is why I tried to keep a reference with

bpy.types.Object.current_draw_manager = draw_callback_handler

that allows to access the draw_callback_handler after the script terminated.
Even with the line above, the example script kept crashing. (However, I'm unsure if bpy.types.Object.current_draw_manager = draw_callback_handler achieves the intended effect.)

I'm marking this as confirmed, but given the reference passing to an object that is no longer valid at a later point in time this might be an unsupported use case / a limitation by the Python API (see the Gotchas of the Python API). However, I believe this is a bug since removing the handler should not be causing the problem.

are going vertex groups, vertex colors etc. eventually replaced with attributes in future? would be nice to have unified access to all data

You'll probably be able to access them as attributes as well (geometry nodes already handles them like "regular" attributes), but the legacy access will stay as well (I dont think this will removed at all)

rBc47990f41c7364058a72f5f162e5cdc06bce0adc